ApeRAG is a production-oriented RAG (Retrieval-Augmented Generation) platform that combines Graph RAG, vector retrieval, and full-text retrieval, and integrates advanced AI agents. It can be used to build knowledge graphs, perform context engineering, and deploy intelligent AI applications capable of autonomously searching and reasoning over knowledge bases.
Key capabilities
- Advanced index types: Provides five index types—vector, full-text, graph, summary, and visual—enabling multidimensional document understanding and retrieval.
- Intelligent AI agents: Built-in AI agents that support MCP (Model Context Protocol) tools, can automatically identify relevant collections, intelligently search content, and provide web search capabilities to complete comprehensive question answering.
- Enhanced Graph RAG and entity normalization: A deeply modified LightRAG implementation with advanced entity normalization (entity merging), generating clearer knowledge graphs and improving relationship understanding.
- Multimodal processing and visual support: Complete multimodal document processing, including visual analysis capabilities for images, charts, and visual content, in parallel with text processing.
- Hybrid retrieval engine: Combines Graph RAG, vector retrieval, full-text retrieval, summary-based retrieval, and vision-based retrieval to achieve comprehensive document understanding.
- MinerU integration: An advanced document parsing service powered by MinerU technology, providing better parsing for complex documents, tables, formulas, and scientific content, with optional GPU acceleration.
- Production-grade deployment: Full Kubernetes support, providing Helm charts and KubeBlocks integration, simplifying the deployment of production-grade databases (PostgreSQL, Redis, Qdrant, Elasticsearch, Neo4j).
- Enterprise management: Built-in audit logs, LLM model management, graph visualization, a comprehensive document management interface, and agent workflow management.
- MCP integration: Full support for the Model Context Protocol (MCP), enabling seamless integration with AI assistants and tools, direct access to knowledge bases, and intelligent querying.
- Developer-friendly: FastAPI backend, React frontend, Celery-based asynchronous task processing, extensive tests, comprehensive development guides, and an agent development framework, making contribution and customization easy.
Usage Docker Compose startup (the simplest way): Ensure Docker and Docker Compose are installed on the machine. After running the corresponding commands, you can access the web interface in the browser http://localhost:3000/web/ and API documentation http://localhost:8000/docs。也可使用 Makefile shortcut (requires GNU Make).
MCP integration: After starting the service, configure it in the MCP client. Authentication methods in order of priority: HTTP Authorization Header (recommended) Authorization: Bearer your-api-key; environment variable (fallback) APERAG_API_KEY=your-api-key. For non-local deployments, use the actual API origin (e.g., https://your-host/mcp/），并将 Replace your-api-key-here with a valid API key in the ApeRAG settings. The MCP server provides: collection browsing (list and explore knowledge collections), hybrid search (search using vector, full-text, and graph methods), and intelligent querying (ask natural language questions about documents).
Kubernetes deployment (recommended for production): requires a Kubernetes cluster (v1.20+), kubectl configured and connected to the cluster, and Helm v3+ installed. First clone the repository to obtain the deployment files. Step 1: Deploy the database services. ApeRAG requires PostgreSQL, Redis, Qdrant, and Elasticsearch. You can choose to use existing databases (edit the connection information in deploy/aperag/values.yaml) or use KubeBlocks for automatic deployment (database connections are preconfigured). Wait for all database Pods to be in the Running state. Step 2: Deploy the ApeRAG application. Configuration options: by default, the doc-ray service is included (requires 4+ CPU cores and 8GB+ memory), and it can be disabled via docray.enabled: false in values.yaml; for more configuration (images, resources, Ingress, etc.), see values.yaml. After deployment, access it via port forwarding. In production, you can configure Ingress in values.yaml for external access.
Use cases Suitable for scenarios that require building your own knowledge graph, performing context engineering, and deploying intelligent AI agents that can autonomously search and reason over a knowledge base. It is especially suitable for teams that need hybrid retrieval, multimodal document processing, enterprise-level management features, and production-grade high-availability deployment.
Limitations and notes Minimum system requirements: must meet the conditions for installing Docker and Docker Compose. Kubernetes deployment must meet the above cluster, kubectl, and Helm version requirements. The doc-ray service included by default has resource requirements (4+ CPU cores, 8GB+ memory). MinerU advanced document parsing is optional, and GPU acceleration is also optional. Other features, fees, prices, platforms, or performance information not explicitly provided in the materials are not yet available.
